[linux.git] / net / rds / af_rds.c

/*
 * Copyright (c) 2006 Oracle.  All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 */
#include <linux/module.h>
#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/gfp.h>
#include <linux/in.h>
#include <linux/poll.h>
#include <net/sock.h>

#include "rds.h"

/* this is just used for stats gathering :/ */
static DEFINE_SPINLOCK(rds_sock_lock);
static unsigned long rds_sock_count;
static LIST_HEAD(rds_sock_list);
DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq);

/*
 * This is called as the final descriptor referencing this socket is closed.
 * We have to unbind the socket so that another socket can be bound to the
 * address it was using.
 *
 * We have to be careful about racing with the incoming path.  sock_orphan()
 * sets SOCK_DEAD and we use that as an indicator to the rx path that new
 * messages shouldn't be queued.
 */
static int rds_release(struct socket *sock)
{
	struct sock *sk = sock->sk;
	struct rds_sock *rs;

	if (!sk)
		goto out;

	rs = rds_sk_to_rs(sk);

	sock_orphan(sk);
	/* Note - rds_clear_recv_queue grabs rs_recv_lock, so
	 * that ensures the recv path has completed messing
	 * with the socket. */
	rds_clear_recv_queue(rs);
	rds_cong_remove_socket(rs);

	rds_remove_bound(rs);

	rds_send_drop_to(rs, NULL);
	rds_rdma_drop_keys(rs);
	rds_notify_queue_get(rs, NULL);

	spin_lock_bh(&rds_sock_lock);
	list_del_init(&rs->rs_item);
	rds_sock_count--;
	spin_unlock_bh(&rds_sock_lock);

	rds_trans_put(rs->rs_transport);

	sock->sk = NULL;
	sock_put(sk);
out:
	return 0;
}

/*
 * Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
 * _bh() isn't OK here, we're called from interrupt handlers.  It's probably OK
 * to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
 * this seems more conservative.
 * NB - normally, one would use sk_callback_lock for this, but we can
 * get here from interrupts, whereas the network code grabs sk_callback_lock
 * with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
 */
void rds_wake_sk_sleep(struct rds_sock *rs)
{
	unsigned long flags;

	read_lock_irqsave(&rs->rs_recv_lock, flags);
	__rds_wake_sk_sleep(rds_rs_to_sk(rs));
	read_unlock_irqrestore(&rs->rs_recv_lock, flags);
}

static int rds_getname(struct socket *sock, struct sockaddr *uaddr,
		       int *uaddr_len, int peer)
{
	struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
	struct rds_sock *rs = rds_sk_to_rs(sock->sk);

	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));

	/* racey, don't care */
	if (peer) {
		if (!rs->rs_conn_addr)
			return -ENOTCONN;

		sin->sin_port = rs->rs_conn_port;
		sin->sin_addr.s_addr = rs->rs_conn_addr;
	} else {
		sin->sin_port = rs->rs_bound_port;
		sin->sin_addr.s_addr = rs->rs_bound_addr;
	}

	sin->sin_family = AF_INET;

	*uaddr_len = sizeof(*sin);
	return 0;
}

/*
 * RDS' poll is without a doubt the least intuitive part of the interface,
 * as POLLIN and POLLOUT do not behave entirely as you would expect from
 * a network protocol.
 *
 * POLLIN is asserted if
 *  -	there is data on the receive queue.
 *  -	to signal that a previously congested destination may have become
 *	uncongested
 *  -	A notification has been queued to the socket (this can be a congestion
 *	update, or a RDMA completion).
 *
 * POLLOUT is asserted if there is room on the send queue. This does not mean
 * however, that the next sendmsg() call will succeed. If the application tries
 * to send to a congested destination, the system call may still fail (and
 * return ENOBUFS).
 */
static unsigned int rds_poll(struct file *file, struct socket *sock,
			     poll_table *wait)
{
	struct sock *sk = sock->sk;
	struct rds_sock *rs = rds_sk_to_rs(sk);
	unsigned int mask = 0;
	unsigned long flags;

	poll_wait(file, sk_sleep(sk), wait);

	if (rs->rs_seen_congestion)
		poll_wait(file, &rds_poll_waitq, wait);

	read_lock_irqsave(&rs->rs_recv_lock, flags);
	if (!rs->rs_cong_monitor) {
		/* When a congestion map was updated, we signal POLLIN for
		 * "historical" reasons. Applications can also poll for
		 * WRBAND instead. */
		if (rds_cong_updated_since(&rs->rs_cong_track))
			mask |= (POLLIN | POLLRDNORM | POLLWRBAND);
	} else {
		spin_lock(&rs->rs_lock);
		if (rs->rs_cong_notify)
			mask |= (POLLIN | POLLRDNORM);
		spin_unlock(&rs->rs_lock);
	}
	if (!list_empty(&rs->rs_recv_queue) ||
	    !list_empty(&rs->rs_notify_queue))
		mask |= (POLLIN | POLLRDNORM);
	if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
		mask |= (POLLOUT | POLLWRNORM);
	read_unlock_irqrestore(&rs->rs_recv_lock, flags);

	/* clear state any time we wake a seen-congested socket */
	if (mask)
		rs->rs_seen_congestion = 0;

	return mask;
}

static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
{
	return -ENOIOCTLCMD;
}

static int rds_cancel_sent_to(struct rds_sock *rs, char __user *optval,
			      int len)
{
	struct sockaddr_in sin;
	int ret = 0;

	/* racing with another thread binding seems ok here */
	if (rs->rs_bound_addr == 0) {
		ret = -ENOTCONN; /* XXX not a great errno */
		goto out;
	}

	if (len < sizeof(struct sockaddr_in)) {
		ret = -EINVAL;
		goto out;
	}

	if (copy_from_user(&sin, optval, sizeof(sin))) {
		ret = -EFAULT;
		goto out;
	}

	rds_send_drop_to(rs, &sin);
out:
	return ret;
}

static int rds_set_bool_option(unsigned char *optvar, char __user *optval,
			       int optlen)
{
	int value;

	if (optlen < sizeof(int))
		return -EINVAL;
	if (get_user(value, (int __user *) optval))
		return -EFAULT;
	*optvar = !!value;
	return 0;
}

static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
			    int optlen)
{
	int ret;

	ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
	if (ret == 0) {
		if (rs->rs_cong_monitor) {
			rds_cong_add_socket(rs);
		} else {
			rds_cong_remove_socket(rs);
			rs->rs_cong_mask = 0;
			rs->rs_cong_notify = 0;
		}
	}
	return ret;
}

static int rds_set_transport(struct rds_sock *rs, char __user *optval,
			     int optlen)
{
	int t_type;

	if (rs->rs_transport)
		return -EOPNOTSUPP; /* previously attached to transport */

	if (optlen != sizeof(int))
		return -EINVAL;

	if (copy_from_user(&t_type, (int __user *)optval, sizeof(t_type)))
		return -EFAULT;

	if (t_type < 0 || t_type >= RDS_TRANS_COUNT)
		return -EINVAL;

	rs->rs_transport = rds_trans_get(t_type);

	return rs->rs_transport ? 0 : -ENOPROTOOPT;
}

static int rds_enable_recvtstamp(struct sock *sk, char __user *optval,
				 int optlen)
{
	int val, valbool;

	if (optlen != sizeof(int))
		return -EFAULT;

	if (get_user(val, (int __user *)optval))
		return -EFAULT;

	valbool = val ? 1 : 0;

	if (valbool)
		sock_set_flag(sk, SOCK_RCVTSTAMP);
	else
		sock_reset_flag(sk, SOCK_RCVTSTAMP);

	return 0;
}

static int rds_recv_track_latency(struct rds_sock *rs, char __user *optval,
				  int optlen)
{
	struct rds_rx_trace_so trace;
	int i;

	if (optlen != sizeof(struct rds_rx_trace_so))
		return -EFAULT;

	if (copy_from_user(&trace, optval, sizeof(trace)))
		return -EFAULT;

	if (trace.rx_traces > RDS_MSG_RX_DGRAM_TRACE_MAX)
		return -EFAULT;

	rs->rs_rx_traces = trace.rx_traces;
	for (i = 0; i < rs->rs_rx_traces; i++) {
		if (trace.rx_trace_pos[i] > RDS_MSG_RX_DGRAM_TRACE_MAX) {
			rs->rs_rx_traces = 0;
			return -EFAULT;
		}
		rs->rs_rx_trace[i] = trace.rx_trace_pos[i];
	}

	return 0;
}

static int rds_setsockopt(struct socket *sock, int level, int optname,
			  char __user *optval, unsigned int optlen)
{
	struct rds_sock *rs = rds_sk_to_rs(sock->sk);
	int ret;

	if (level != SOL_RDS) {
		ret = -ENOPROTOOPT;
		goto out;
	}

	switch (optname) {
	case RDS_CANCEL_SENT_TO:
		ret = rds_cancel_sent_to(rs, optval, optlen);
		break;
	case RDS_GET_MR:
		ret = rds_get_mr(rs, optval, optlen);
		break;
	case RDS_GET_MR_FOR_DEST:
		ret = rds_get_mr_for_dest(rs, optval, optlen);
		break;
	case RDS_FREE_MR:
		ret = rds_free_mr(rs, optval, optlen);
		break;
	case RDS_RECVERR:
		ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
		break;
	case RDS_CONG_MONITOR:
		ret = rds_cong_monitor(rs, optval, optlen);
		break;
	case SO_RDS_TRANSPORT:
		lock_sock(sock->sk);
		ret = rds_set_transport(rs, optval, optlen);
		release_sock(sock->sk);
		break;
	case SO_TIMESTAMP:
		lock_sock(sock->sk);
		ret = rds_enable_recvtstamp(sock->sk, optval, optlen);
		release_sock(sock->sk);
		break;
	case SO_RDS_MSG_RXPATH_LATENCY:
		ret = rds_recv_track_latency(rs, optval, optlen);
		break;
	default:
		ret = -ENOPROTOOPT;
	}
out:
	return ret;
}

static int rds_getsockopt(struct socket *sock, int level, int optname,
			  char __user *optval, int __user *optlen)
{
	struct rds_sock *rs = rds_sk_to_rs(sock->sk);
	int ret = -ENOPROTOOPT, len;
	int trans;

	if (level != SOL_RDS)
		goto out;

	if (get_user(len, optlen)) {
		ret = -EFAULT;
		goto out;
	}

	switch (optname) {
	case RDS_INFO_FIRST ... RDS_INFO_LAST:
		ret = rds_info_getsockopt(sock, optname, optval,
					  optlen);
		break;

	case RDS_RECVERR:
		if (len < sizeof(int))
			ret = -EINVAL;
		else
		if (put_user(rs->rs_recverr, (int __user *) optval) ||
		    put_user(sizeof(int), optlen))
			ret = -EFAULT;
		else
			ret = 0;
		break;
	case SO_RDS_TRANSPORT:
		if (len < sizeof(int)) {
			ret = -EINVAL;
			break;
		}
		trans = (rs->rs_transport ? rs->rs_transport->t_type :
			 RDS_TRANS_NONE); /* unbound */
		if (put_user(trans, (int __user *)optval) ||
		    put_user(sizeof(int), optlen))
			ret = -EFAULT;
		else
			ret = 0;
		break;
	default:
		break;
	}

out:
	return ret;

}

static int rds_connect(struct socket *sock, struct sockaddr *uaddr,
		       int addr_len, int flags)
{
	struct sock *sk = sock->sk;
	struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
	struct rds_sock *rs = rds_sk_to_rs(sk);
	int ret = 0;

	lock_sock(sk);

	if (addr_len != sizeof(struct sockaddr_in)) {
		ret = -EINVAL;
		goto out;
	}

	if (sin->sin_family != AF_INET) {
		ret = -EAFNOSUPPORT;
		goto out;
	}

	if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
		ret = -EDESTADDRREQ;
		goto out;
	}

	rs->rs_conn_addr = sin->sin_addr.s_addr;
	rs->rs_conn_port = sin->sin_port;

out:
	release_sock(sk);
	return ret;
}

static struct proto rds_proto = {
	.name	  = "RDS",
	.owner	  = THIS_MODULE,
	.obj_size = sizeof(struct rds_sock),
};

static const struct proto_ops rds_proto_ops = {
	.family =	AF_RDS,
	.owner =	THIS_MODULE,
	.release =	rds_release,
	.bind =		rds_bind,
	.connect =	rds_connect,
	.socketpair =	sock_no_socketpair,
	.accept =	sock_no_accept,
	.getname =	rds_getname,
	.poll =		rds_poll,
	.ioctl =	rds_ioctl,
	.listen =	sock_no_listen,
	.shutdown =	sock_no_shutdown,
	.setsockopt =	rds_setsockopt,
	.getsockopt =	rds_getsockopt,
	.sendmsg =	rds_sendmsg,
	.recvmsg =	rds_recvmsg,
	.mmap =		sock_no_mmap,
	.sendpage =	sock_no_sendpage,
};

static void rds_sock_destruct(struct sock *sk)
{
	struct rds_sock *rs = rds_sk_to_rs(sk);

	WARN_ON((&rs->rs_item != rs->rs_item.next ||
		 &rs->rs_item != rs->rs_item.prev));
}

static int __rds_create(struct socket *sock, struct sock *sk, int protocol)
{
	struct rds_sock *rs;

	sock_init_data(sock, sk);
	sock->ops		= &rds_proto_ops;
	sk->sk_protocol		= protocol;
	sk->sk_destruct		= rds_sock_destruct;

	rs = rds_sk_to_rs(sk);
	spin_lock_init(&rs->rs_lock);
	rwlock_init(&rs->rs_recv_lock);
	INIT_LIST_HEAD(&rs->rs_send_queue);
	INIT_LIST_HEAD(&rs->rs_recv_queue);
	INIT_LIST_HEAD(&rs->rs_notify_queue);
	INIT_LIST_HEAD(&rs->rs_cong_list);
	spin_lock_init(&rs->rs_rdma_lock);
	rs->rs_rdma_keys = RB_ROOT;
	rs->rs_rx_traces = 0;

	spin_lock_bh(&rds_sock_lock);
	list_add_tail(&rs->rs_item, &rds_sock_list);
	rds_sock_count++;
	spin_unlock_bh(&rds_sock_lock);

	return 0;
}

static int rds_create(struct net *net, struct socket *sock, int protocol,
		      int kern)
{
	struct sock *sk;

	if (sock->type != SOCK_SEQPACKET || protocol)
		return -ESOCKTNOSUPPORT;

	sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto, kern);
	if (!sk)
		return -ENOMEM;

	return __rds_create(sock, sk, protocol);
}

void rds_sock_addref(struct rds_sock *rs)
{
	sock_hold(rds_rs_to_sk(rs));
}

void rds_sock_put(struct rds_sock *rs)
{
	sock_put(rds_rs_to_sk(rs));
}

static const struct net_proto_family rds_family_ops = {
	.family =	AF_RDS,
	.create =	rds_create,
	.owner	=	THIS_MODULE,
};

static void rds_sock_inc_info(struct socket *sock, unsigned int len,
			      struct rds_info_iterator *iter,
			      struct rds_info_lengths *lens)
{
	struct rds_sock *rs;
	struct rds_incoming *inc;
	unsigned int total = 0;

	len /= sizeof(struct rds_info_message);

	spin_lock_bh(&rds_sock_lock);

	list_for_each_entry(rs, &rds_sock_list, rs_item) {
		read_lock(&rs->rs_recv_lock);

		/* XXX too lazy to maintain counts.. */
		list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
			total++;
			if (total <= len)
				rds_inc_info_copy(inc, iter, inc->i_saddr,
						  rs->rs_bound_addr, 1);
		}

		read_unlock(&rs->rs_recv_lock);
	}

	spin_unlock_bh(&rds_sock_lock);

	lens->nr = total;
	lens->each = sizeof(struct rds_info_message);
}

static void rds_sock_info(struct socket *sock, unsigned int len,
			  struct rds_info_iterator *iter,
			  struct rds_info_lengths *lens)
{
	struct rds_info_socket sinfo;
	struct rds_sock *rs;

	len /= sizeof(struct rds_info_socket);

	spin_lock_bh(&rds_sock_lock);

	if (len < rds_sock_count)
		goto out;

	list_for_each_entry(rs, &rds_sock_list, rs_item) {
		sinfo.sndbuf = rds_sk_sndbuf(rs);
		sinfo.rcvbuf = rds_sk_rcvbuf(rs);
		sinfo.bound_addr = rs->rs_bound_addr;
		sinfo.connected_addr = rs->rs_conn_addr;
		sinfo.bound_port = rs->rs_bound_port;
		sinfo.connected_port = rs->rs_conn_port;
		sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));

		rds_info_copy(iter, &sinfo, sizeof(sinfo));
	}

out:
	lens->nr = rds_sock_count;
	lens->each = sizeof(struct rds_info_socket);

	spin_unlock_bh(&rds_sock_lock);
}

static void rds_exit(void)
{
	sock_unregister(rds_family_ops.family);
	proto_unregister(&rds_proto);
	rds_conn_exit();
	rds_cong_exit();
	rds_sysctl_exit();
	rds_threads_exit();
	rds_stats_exit();
	rds_page_exit();
	rds_bind_lock_destroy();
	rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
	rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
}
module_exit(rds_exit);

u32 rds_gen_num;

static int rds_init(void)
{
	int ret;

	net_get_random_once(&rds_gen_num, sizeof(rds_gen_num));

	ret = rds_bind_lock_init();
	if (ret)
		goto out;

	ret = rds_conn_init();
	if (ret)
		goto out_bind;

	ret = rds_threads_init();
	if (ret)
		goto out_conn;
	ret = rds_sysctl_init();
	if (ret)
		goto out_threads;
	ret = rds_stats_init();
	if (ret)
		goto out_sysctl;
	ret = proto_register(&rds_proto, 1);
	if (ret)
		goto out_stats;
	ret = sock_register(&rds_family_ops);
	if (ret)
		goto out_proto;

	rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
	rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);

	goto out;

out_proto:
	proto_unregister(&rds_proto);
out_stats:
	rds_stats_exit();
out_sysctl:
	rds_sysctl_exit();
out_threads:
	rds_threads_exit();
out_conn:
	rds_conn_exit();
	rds_cong_exit();
	rds_page_exit();
out_bind:
	rds_bind_lock_destroy();
out:
	return ret;
}
module_init(rds_init);

#define DRV_VERSION     "4.0"
#define DRV_RELDATE     "Feb 12, 2009"

MODULE_AUTHOR("Oracle Corporation <[email protected]>");
MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
		   " v" DRV_VERSION " (" DRV_RELDATE ")");
MODULE_VERSION(DRV_VERSION);
MODULE_LICENSE("Dual BSD/GPL");
MODULE_ALIAS_NETPROTO(PF_RDS);
Commit	Line	Data
639b321b AG	1	/*
	2	* Copyright (c) 2006 Oracle. All rights reserved.
	3	*
	4	* This software is available to you under a choice of one of two
	5	* licenses. You may choose to be licensed under the terms of the GNU
	6	* General Public License (GPL) Version 2, available from the file
	7	* COPYING in the main directory of this source tree, or the
	8	* OpenIB.org BSD license below:
	9	*
	10	* Redistribution and use in source and binary forms, with or
	11	* without modification, are permitted provided that the following
	12	* conditions are met:
	13	*
	14	* - Redistributions of source code must retain the above
	15	* copyright notice, this list of conditions and the following
	16	* disclaimer.
	17	*
	18	* - Redistributions in binary form must reproduce the above
	19	* copyright notice, this list of conditions and the following
	20	* disclaimer in the documentation and/or other materials
	21	* provided with the distribution.
	22	*
	23	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	24	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	25	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	26	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	27	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	28	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	29	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	30	* SOFTWARE.
	31	*
	32	*/
	33	#include <linux/module.h>
	34	#include <linux/errno.h>
	35	#include <linux/kernel.h>
5a0e3ad6	36	#include <linux/gfp.h>
639b321b AG	37	#include <linux/in.h>
639b321b AG	38	#include <linux/poll.h>
639b321b AG	39	#include <net/sock.h>
	40
	41	#include "rds.h"
639b321b AG	42
	43	/* this is just used for stats gathering :/ */
	44	static DEFINE_SPINLOCK(rds_sock_lock);
	45	static unsigned long rds_sock_count;
	46	static LIST_HEAD(rds_sock_list);
	47	DECLARE_WAIT_QUEUE_HEAD(rds_poll_waitq);
	48
	49	/*
	50	* This is called as the final descriptor referencing this socket is closed.
	51	* We have to unbind the socket so that another socket can be bound to the
	52	* address it was using.
	53	*
	54	* We have to be careful about racing with the incoming path. sock_orphan()
	55	* sets SOCK_DEAD and we use that as an indicator to the rx path that new
	56	* messages shouldn't be queued.
	57	*/
	58	static int rds_release(struct socket *sock)
	59	{
	60	struct sock *sk = sock->sk;
	61	struct rds_sock *rs;
639b321b	62
8690bfa1	63	if (!sk)
639b321b AG	64	goto out;
	65
	66	rs = rds_sk_to_rs(sk);
	67
	68	sock_orphan(sk);
	69	/* Note - rds_clear_recv_queue grabs rs_recv_lock, so
	70	* that ensures the recv path has completed messing
	71	* with the socket. */
	72	rds_clear_recv_queue(rs);
	73	rds_cong_remove_socket(rs);
38a4e5e6	74
639b321b	75	rds_remove_bound(rs);
38a4e5e6	76
639b321b AG	77	rds_send_drop_to(rs, NULL);
	78	rds_rdma_drop_keys(rs);
	79	rds_notify_queue_get(rs, NULL);
	80
efc3dbc3	81	spin_lock_bh(&rds_sock_lock);
639b321b AG	82	list_del_init(&rs->rs_item);
639b321b AG	83	rds_sock_count--;
efc3dbc3	84	spin_unlock_bh(&rds_sock_lock);
639b321b	85
5adb5bc6 ZB	86	rds_trans_put(rs->rs_transport);
5adb5bc6 ZB	87
639b321b AG	88	sock->sk = NULL;
	89	sock_put(sk);
	90	out:
	91	return 0;
	92	}
	93
	94	/*
	95	* Careful not to race with rds_release -> sock_orphan which clears sk_sleep.
	96	* _bh() isn't OK here, we're called from interrupt handlers. It's probably OK
	97	* to wake the waitqueue after sk_sleep is clear as we hold a sock ref, but
	98	* this seems more conservative.
	99	* NB - normally, one would use sk_callback_lock for this, but we can
	100	* get here from interrupts, whereas the network code grabs sk_callback_lock
	101	* with _lock_bh only - so relying on sk_callback_lock introduces livelocks.
	102	*/
	103	void rds_wake_sk_sleep(struct rds_sock *rs)
	104	{
	105	unsigned long flags;
	106
	107	read_lock_irqsave(&rs->rs_recv_lock, flags);
	108	__rds_wake_sk_sleep(rds_rs_to_sk(rs));
	109	read_unlock_irqrestore(&rs->rs_recv_lock, flags);
	110	}
	111
	112	static int rds_getname(struct socket sock, struct sockaddr uaddr,
	113	int *uaddr_len, int peer)
	114	{
	115	struct sockaddr_in sin = (struct sockaddr_in )uaddr;
	116	struct rds_sock *rs = rds_sk_to_rs(sock->sk);
	117
	118	memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
	119
	120	/* racey, don't care */
	121	if (peer) {
	122	if (!rs->rs_conn_addr)
	123	return -ENOTCONN;
	124
	125	sin->sin_port = rs->rs_conn_port;
	126	sin->sin_addr.s_addr = rs->rs_conn_addr;
	127	} else {
	128	sin->sin_port = rs->rs_bound_port;
	129	sin->sin_addr.s_addr = rs->rs_bound_addr;
	130	}
	131
	132	sin->sin_family = AF_INET;
	133
	134	uaddr_len = sizeof(sin);
	135	return 0;
	136	}
	137
	138	/*
	139	* RDS' poll is without a doubt the least intuitive part of the interface,
	140	* as POLLIN and POLLOUT do not behave entirely as you would expect from
	141	* a network protocol.
	142	*
	143	* POLLIN is asserted if
	144	* - there is data on the receive queue.
	145	* - to signal that a previously congested destination may have become
	146	* uncongested
	147	* - A notification has been queued to the socket (this can be a congestion
	148	* update, or a RDMA completion).
	149	*
	150	* POLLOUT is asserted if there is room on the send queue. This does not mean
	151	* however, that the next sendmsg() call will succeed. If the application tries
152	* to send to a congested destination, the system call may still fail (and
153	* return ENOBUFS).
154	*/
155	static unsigned int rds_poll(struct file file, struct socket sock,
156	poll_table *wait)
157	{
158	struct sock *sk = sock->sk;
159	struct rds_sock *rs = rds_sk_to_rs(sk);
160	unsigned int mask = 0;
161	unsigned long flags;
162
aa395145	163	poll_wait(file, sk_sleep(sk), wait);
639b321b	164
b98ba52f AG	165	if (rs->rs_seen_congestion)
b98ba52f AG	166	poll_wait(file, &rds_poll_waitq, wait);
639b321b AG	167
	168	read_lock_irqsave(&rs->rs_recv_lock, flags);
	169	if (!rs->rs_cong_monitor) {
	170	/* When a congestion map was updated, we signal POLLIN for
	171	* "historical" reasons. Applications can also poll for
	172	* WRBAND instead. */
	173	if (rds_cong_updated_since(&rs->rs_cong_track))
	174	mask \|= (POLLIN \| POLLRDNORM \| POLLWRBAND);
	175	} else {
	176	spin_lock(&rs->rs_lock);
	177	if (rs->rs_cong_notify)
	178	mask \|= (POLLIN \| POLLRDNORM);
	179	spin_unlock(&rs->rs_lock);
	180	}
f64f9e71 JP	181	if (!list_empty(&rs->rs_recv_queue) \|\|
f64f9e71 JP	182	!list_empty(&rs->rs_notify_queue))
639b321b AG	183	mask \|= (POLLIN \| POLLRDNORM);
	184	if (rs->rs_snd_bytes < rds_sk_sndbuf(rs))
	185	mask \|= (POLLOUT \| POLLWRNORM);
	186	read_unlock_irqrestore(&rs->rs_recv_lock, flags);
	187
b98ba52f AG	188	/* clear state any time we wake a seen-congested socket */
	189	if (mask)
	190	rs->rs_seen_congestion = 0;
	191
639b321b AG	192	return mask;
	193	}
	194
	195	static int rds_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
	196	{
	197	return -ENOIOCTLCMD;
	198	}
	199
	200	static int rds_cancel_sent_to(struct rds_sock rs, char __user optval,
	201	int len)
	202	{
	203	struct sockaddr_in sin;
	204	int ret = 0;
	205
	206	/* racing with another thread binding seems ok here */
	207	if (rs->rs_bound_addr == 0) {
	208	ret = -ENOTCONN; /* XXX not a great errno */
	209	goto out;
	210	}
	211
	212	if (len < sizeof(struct sockaddr_in)) {
	213	ret = -EINVAL;
	214	goto out;
	215	}
	216
	217	if (copy_from_user(&sin, optval, sizeof(sin))) {
	218	ret = -EFAULT;
	219	goto out;
	220	}
	221
	222	rds_send_drop_to(rs, &sin);
	223	out:
	224	return ret;
	225	}
	226
	227	static int rds_set_bool_option(unsigned char optvar, char __user optval,
	228	int optlen)
	229	{
	230	int value;
	231
	232	if (optlen < sizeof(int))
	233	return -EINVAL;
	234	if (get_user(value, (int __user *) optval))
	235	return -EFAULT;
	236	*optvar = !!value;
	237	return 0;
	238	}
	239
	240	static int rds_cong_monitor(struct rds_sock rs, char __user optval,
	241	int optlen)
	242	{
	243	int ret;
	244
	245	ret = rds_set_bool_option(&rs->rs_cong_monitor, optval, optlen);
	246	if (ret == 0) {
	247	if (rs->rs_cong_monitor) {
	248	rds_cong_add_socket(rs);
	249	} else {
	250	rds_cong_remove_socket(rs);
	251	rs->rs_cong_mask = 0;
	252	rs->rs_cong_notify = 0;
	253	}
	254	}
	255	return ret;
256	}
257
d97dac54 SV	258	static int rds_set_transport(struct rds_sock rs, char __user optval,
	259	int optlen)
	260	{
	261	int t_type;
	262
	263	if (rs->rs_transport)
	264	return -EOPNOTSUPP; /* previously attached to transport */
	265
	266	if (optlen != sizeof(int))
	267	return -EINVAL;
	268
	269	if (copy_from_user(&t_type, (int __user *)optval, sizeof(t_type)))
	270	return -EFAULT;
	271
	272	if (t_type < 0 \|\| t_type >= RDS_TRANS_COUNT)
	273	return -EINVAL;
	274
	275	rs->rs_transport = rds_trans_get(t_type);
	276
	277	return rs->rs_transport ? 0 : -ENOPROTOOPT;
	278	}
	279
5711f8b3	280	static int rds_enable_recvtstamp(struct sock sk, char __user optval,
	281	int optlen)
	282	{
	283	int val, valbool;
	284
	285	if (optlen != sizeof(int))
	286	return -EFAULT;
	287
	288	if (get_user(val, (int __user *)optval))
	289	return -EFAULT;
	290
	291	valbool = val ? 1 : 0;
	292
	293	if (valbool)
	294	sock_set_flag(sk, SOCK_RCVTSTAMP);
	295	else
	296	sock_reset_flag(sk, SOCK_RCVTSTAMP);
	297
	298	return 0;
	299	}
	300
3289025a SS	301	static int rds_recv_track_latency(struct rds_sock rs, char __user optval,
	302	int optlen)
	303	{
	304	struct rds_rx_trace_so trace;
	305	int i;
	306
	307	if (optlen != sizeof(struct rds_rx_trace_so))
	308	return -EFAULT;
	309
	310	if (copy_from_user(&trace, optval, sizeof(trace)))
	311	return -EFAULT;
	312
780e9829	313	if (trace.rx_traces > RDS_MSG_RX_DGRAM_TRACE_MAX)
	314	return -EFAULT;
	315
3289025a SS	316	rs->rs_rx_traces = trace.rx_traces;
	317	for (i = 0; i < rs->rs_rx_traces; i++) {
	318	if (trace.rx_trace_pos[i] > RDS_MSG_RX_DGRAM_TRACE_MAX) {
	319	rs->rs_rx_traces = 0;
	320	return -EFAULT;
	321	}
	322	rs->rs_rx_trace[i] = trace.rx_trace_pos[i];
	323	}
	324
	325	return 0;
	326	}
	327
639b321b	328	static int rds_setsockopt(struct socket *sock, int level, int optname,
b7058842	329	char __user *optval, unsigned int optlen)
639b321b AG	330	{
	331	struct rds_sock *rs = rds_sk_to_rs(sock->sk);
	332	int ret;
	333
	334	if (level != SOL_RDS) {
	335	ret = -ENOPROTOOPT;
	336	goto out;
	337	}
	338
	339	switch (optname) {
	340	case RDS_CANCEL_SENT_TO:
	341	ret = rds_cancel_sent_to(rs, optval, optlen);
	342	break;
	343	case RDS_GET_MR:
	344	ret = rds_get_mr(rs, optval, optlen);
	345	break;
244546f0 AG	346	case RDS_GET_MR_FOR_DEST:
	347	ret = rds_get_mr_for_dest(rs, optval, optlen);
	348	break;
639b321b AG	349	case RDS_FREE_MR:
	350	ret = rds_free_mr(rs, optval, optlen);
	351	break;
	352	case RDS_RECVERR:
	353	ret = rds_set_bool_option(&rs->rs_recverr, optval, optlen);
	354	break;
	355	case RDS_CONG_MONITOR:
	356	ret = rds_cong_monitor(rs, optval, optlen);
	357	break;
d97dac54 SV	358	case SO_RDS_TRANSPORT:
	359	lock_sock(sock->sk);
	360	ret = rds_set_transport(rs, optval, optlen);
	361	release_sock(sock->sk);
	362	break;
5711f8b3	363	case SO_TIMESTAMP:
	364	lock_sock(sock->sk);
	365	ret = rds_enable_recvtstamp(sock->sk, optval, optlen);
	366	release_sock(sock->sk);
	367	break;
3289025a SS	368	case SO_RDS_MSG_RXPATH_LATENCY:
	369	ret = rds_recv_track_latency(rs, optval, optlen);
	370	break;
639b321b AG	371	default:
	372	ret = -ENOPROTOOPT;
	373	}
	374	out:
	375	return ret;
	376	}
	377
	378	static int rds_getsockopt(struct socket *sock, int level, int optname,
	379	char __user optval, int __user optlen)
	380	{
	381	struct rds_sock *rs = rds_sk_to_rs(sock->sk);
	382	int ret = -ENOPROTOOPT, len;
8ba38460	383	int trans;
639b321b AG	384
	385	if (level != SOL_RDS)
	386	goto out;
	387
	388	if (get_user(len, optlen)) {
	389	ret = -EFAULT;
	390	goto out;
	391	}
	392
	393	switch (optname) {
	394	case RDS_INFO_FIRST ... RDS_INFO_LAST:
	395	ret = rds_info_getsockopt(sock, optname, optval,
	396	optlen);
	397	break;
	398
	399	case RDS_RECVERR:
	400	if (len < sizeof(int))
	401	ret = -EINVAL;
	402	else
f64f9e71 JP	403	if (put_user(rs->rs_recverr, (int __user *) optval) \|\|
f64f9e71 JP	404	put_user(sizeof(int), optlen))
639b321b AG	405	ret = -EFAULT;
	406	else
	407	ret = 0;
	408	break;
8ba38460 SV	409	case SO_RDS_TRANSPORT:
	410	if (len < sizeof(int)) {
	411	ret = -EINVAL;
	412	break;
	413	}
	414	trans = (rs->rs_transport ? rs->rs_transport->t_type :
	415	RDS_TRANS_NONE); /* unbound */
	416	if (put_user(trans, (int __user *)optval) \|\|
	417	put_user(sizeof(int), optlen))
	418	ret = -EFAULT;
	419	else
	420	ret = 0;
	421	break;
639b321b AG	422	default:
	423	break;
	424	}
	425
	426	out:
	427	return ret;
	428
	429	}
	430
	431	static int rds_connect(struct socket sock, struct sockaddr uaddr,
	432	int addr_len, int flags)
	433	{
	434	struct sock *sk = sock->sk;
	435	struct sockaddr_in sin = (struct sockaddr_in )uaddr;
	436	struct rds_sock *rs = rds_sk_to_rs(sk);
	437	int ret = 0;
	438
	439	lock_sock(sk);
	440
	441	if (addr_len != sizeof(struct sockaddr_in)) {
	442	ret = -EINVAL;
	443	goto out;
	444	}
	445
	446	if (sin->sin_family != AF_INET) {
	447	ret = -EAFNOSUPPORT;
	448	goto out;
	449	}
	450
	451	if (sin->sin_addr.s_addr == htonl(INADDR_ANY)) {
	452	ret = -EDESTADDRREQ;
	453	goto out;
	454	}
	455
	456	rs->rs_conn_addr = sin->sin_addr.s_addr;
	457	rs->rs_conn_port = sin->sin_port;
	458
	459	out:
	460	release_sock(sk);
	461	return ret;
	462	}
	463
	464	static struct proto rds_proto = {
	465	.name = "RDS",
	466	.owner = THIS_MODULE,
	467	.obj_size = sizeof(struct rds_sock),
	468	};
	469
5708e868	470	static const struct proto_ops rds_proto_ops = {
639b321b AG	471	.family = AF_RDS,
	472	.owner = THIS_MODULE,
	473	.release = rds_release,
	474	.bind = rds_bind,
	475	.connect = rds_connect,
	476	.socketpair = sock_no_socketpair,
	477	.accept = sock_no_accept,
	478	.getname = rds_getname,
	479	.poll = rds_poll,
	480	.ioctl = rds_ioctl,
	481	.listen = sock_no_listen,
	482	.shutdown = sock_no_shutdown,
	483	.setsockopt = rds_setsockopt,
	484	.getsockopt = rds_getsockopt,
	485	.sendmsg = rds_sendmsg,
	486	.recvmsg = rds_recvmsg,
	487	.mmap = sock_no_mmap,
	488	.sendpage = sock_no_sendpage,
	489	};
	490
0df5f9a6	491	static void rds_sock_destruct(struct sock *sk)
	492	{
	493	struct rds_sock *rs = rds_sk_to_rs(sk);
	494
	495	WARN_ON((&rs->rs_item != rs->rs_item.next \|\|
	496	&rs->rs_item != rs->rs_item.prev));
	497	}
	498
639b321b AG	499	static int __rds_create(struct socket sock, struct sock sk, int protocol)
639b321b AG	500	{
639b321b AG	501	struct rds_sock *rs;
	502
	503	sock_init_data(sock, sk);
	504	sock->ops = &rds_proto_ops;
	505	sk->sk_protocol = protocol;
0df5f9a6	506	sk->sk_destruct = rds_sock_destruct;
639b321b AG	507
	508	rs = rds_sk_to_rs(sk);
	509	spin_lock_init(&rs->rs_lock);
	510	rwlock_init(&rs->rs_recv_lock);
	511	INIT_LIST_HEAD(&rs->rs_send_queue);
	512	INIT_LIST_HEAD(&rs->rs_recv_queue);
	513	INIT_LIST_HEAD(&rs->rs_notify_queue);
	514	INIT_LIST_HEAD(&rs->rs_cong_list);
	515	spin_lock_init(&rs->rs_rdma_lock);
	516	rs->rs_rdma_keys = RB_ROOT;
3289025a	517	rs->rs_rx_traces = 0;
639b321b	518
efc3dbc3	519	spin_lock_bh(&rds_sock_lock);
639b321b AG	520	list_add_tail(&rs->rs_item, &rds_sock_list);
639b321b AG	521	rds_sock_count++;
efc3dbc3	522	spin_unlock_bh(&rds_sock_lock);
639b321b AG	523
	524	return 0;
	525	}
	526
3f378b68 EP	527	static int rds_create(struct net net, struct socket sock, int protocol,
3f378b68 EP	528	int kern)
639b321b AG	529	{
	530	struct sock *sk;
	531
	532	if (sock->type != SOCK_SEQPACKET \|\| protocol)
	533	return -ESOCKTNOSUPPORT;
	534
11aa9c28	535	sk = sk_alloc(net, AF_RDS, GFP_ATOMIC, &rds_proto, kern);
639b321b AG	536	if (!sk)
	537	return -ENOMEM;
	538
	539	return __rds_create(sock, sk, protocol);
	540	}
	541
	542	void rds_sock_addref(struct rds_sock *rs)
	543	{
	544	sock_hold(rds_rs_to_sk(rs));
	545	}
	546
	547	void rds_sock_put(struct rds_sock *rs)
	548	{
	549	sock_put(rds_rs_to_sk(rs));
	550	}
	551
ec1b4cf7	552	static const struct net_proto_family rds_family_ops = {
639b321b AG	553	.family = AF_RDS,
	554	.create = rds_create,
	555	.owner = THIS_MODULE,
	556	};
	557
	558	static void rds_sock_inc_info(struct socket *sock, unsigned int len,
	559	struct rds_info_iterator *iter,
	560	struct rds_info_lengths *lens)
	561	{
	562	struct rds_sock *rs;
639b321b	563	struct rds_incoming *inc;
639b321b AG	564	unsigned int total = 0;
	565
	566	len /= sizeof(struct rds_info_message);
	567
efc3dbc3	568	spin_lock_bh(&rds_sock_lock);
639b321b AG	569
639b321b AG	570	list_for_each_entry(rs, &rds_sock_list, rs_item) {
639b321b AG	571	read_lock(&rs->rs_recv_lock);
	572
	573	/* XXX too lazy to maintain counts.. */
	574	list_for_each_entry(inc, &rs->rs_recv_queue, i_item) {
	575	total++;
	576	if (total <= len)
	577	rds_inc_info_copy(inc, iter, inc->i_saddr,
	578	rs->rs_bound_addr, 1);
	579	}
	580
	581	read_unlock(&rs->rs_recv_lock);
	582	}
	583
efc3dbc3	584	spin_unlock_bh(&rds_sock_lock);
639b321b AG	585
	586	lens->nr = total;
	587	lens->each = sizeof(struct rds_info_message);
	588	}
	589
	590	static void rds_sock_info(struct socket *sock, unsigned int len,
	591	struct rds_info_iterator *iter,
	592	struct rds_info_lengths *lens)
	593	{
	594	struct rds_info_socket sinfo;
	595	struct rds_sock *rs;
639b321b AG	596
	597	len /= sizeof(struct rds_info_socket);
	598
efc3dbc3	599	spin_lock_bh(&rds_sock_lock);
639b321b AG	600
	601	if (len < rds_sock_count)
	602	goto out;
	603
	604	list_for_each_entry(rs, &rds_sock_list, rs_item) {
	605	sinfo.sndbuf = rds_sk_sndbuf(rs);
	606	sinfo.rcvbuf = rds_sk_rcvbuf(rs);
	607	sinfo.bound_addr = rs->rs_bound_addr;
	608	sinfo.connected_addr = rs->rs_conn_addr;
	609	sinfo.bound_port = rs->rs_bound_port;
	610	sinfo.connected_port = rs->rs_conn_port;
	611	sinfo.inum = sock_i_ino(rds_rs_to_sk(rs));
	612
	613	rds_info_copy(iter, &sinfo, sizeof(sinfo));
	614	}
	615
	616	out:
	617	lens->nr = rds_sock_count;
	618	lens->each = sizeof(struct rds_info_socket);
	619
efc3dbc3	620	spin_unlock_bh(&rds_sock_lock);
639b321b AG	621	}
639b321b AG	622
ef87b7ea	623	static void rds_exit(void)
639b321b	624	{
639b321b AG	625	sock_unregister(rds_family_ops.family);
	626	proto_unregister(&rds_proto);
	627	rds_conn_exit();
	628	rds_cong_exit();
	629	rds_sysctl_exit();
	630	rds_threads_exit();
	631	rds_stats_exit();
	632	rds_page_exit();
7b565434	633	rds_bind_lock_destroy();
639b321b AG	634	rds_info_deregister_func(RDS_INFO_SOCKETS, rds_sock_info);
	635	rds_info_deregister_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
	636	}
	637	module_exit(rds_exit);
	638
905dd418 SV	639	u32 rds_gen_num;
905dd418 SV	640
ef87b7ea	641	static int rds_init(void)
639b321b AG	642	{
	643	int ret;
	644
905dd418 SV	645	net_get_random_once(&rds_gen_num, sizeof(rds_gen_num));
905dd418 SV	646
7b565434	647	ret = rds_bind_lock_init();
	648	if (ret)
	649	goto out;
9b9acde7	650
639b321b AG	651	ret = rds_conn_init();
639b321b AG	652	if (ret)
7b565434	653	goto out_bind;
7b565434	654
639b321b AG	655	ret = rds_threads_init();
	656	if (ret)
	657	goto out_conn;
	658	ret = rds_sysctl_init();
	659	if (ret)
	660	goto out_threads;
	661	ret = rds_stats_init();
	662	if (ret)
	663	goto out_sysctl;
	664	ret = proto_register(&rds_proto, 1);
	665	if (ret)
	666	goto out_stats;
	667	ret = sock_register(&rds_family_ops);
	668	if (ret)
	669	goto out_proto;
	670
	671	rds_info_register_func(RDS_INFO_SOCKETS, rds_sock_info);
	672	rds_info_register_func(RDS_INFO_RECV_MESSAGES, rds_sock_inc_info);
	673
639b321b AG	674	goto out;
639b321b AG	675
639b321b AG	676	out_proto:
	677	proto_unregister(&rds_proto);
	678	out_stats:
	679	rds_stats_exit();
	680	out_sysctl:
	681	rds_sysctl_exit();
	682	out_threads:
	683	rds_threads_exit();
	684	out_conn:
	685	rds_conn_exit();
	686	rds_cong_exit();
	687	rds_page_exit();
7b565434	688	out_bind:
7b565434	689	rds_bind_lock_destroy();
639b321b AG	690	out:
	691	return ret;
	692	}
	693	module_init(rds_init);
	694
	695	#define DRV_VERSION "4.0"
	696	#define DRV_RELDATE "Feb 12, 2009"
	697
	698	MODULE_AUTHOR("Oracle Corporation <[email protected]>");
	699	MODULE_DESCRIPTION("RDS: Reliable Datagram Sockets"
	700	" v" DRV_VERSION " (" DRV_RELDATE ")");
	701	MODULE_VERSION(DRV_VERSION);
	702	MODULE_LICENSE("Dual BSD/GPL");
	703	MODULE_ALIAS_NETPROTO(PF_RDS);